At SVP, AI Drives Electricity Demand — and Soon Will Also Manage It

Silicon Valley Power has teamed with Emerald AI to explore how technology can make the best use of the city's existing infrastructure.

Data centers generate 13% of Santa Clara’s general fund revenue and consume 60% of Silicon Valley Power’s (SVP) electricity. It stands to reason that if data centers can reduce power use without performance hits, utilities gain a tool to increase grid reliability and make better use of existing infrastructure.

Long known for innovation, Santa Clara’s city-owned electric utility is out ahead again. SVP is among the first U.S. utilities to implement real-time, adaptive data center power management.

In a pilot project at an Nvidia data center, SVP is using Emerald AI’s Conductor platform to shift, delay, or redistribute computing tasks to reduce power draw in real time — without affecting high-priority jobs.

“All We Have is a Blunt Instrument”

Electrical load monitoring systems aren’t new. But today, if a transmission line goes out of service, “all we have is a blunt instrument” for reducing load, Procos said.

“We have the ability to drop an entire feeder — a circuit that connects to customers — but you might have three or four data centers on that feeder,” he said. “That might be 200 megawatts when we only need to reduce 50.”

Making it harder: “One of our people would have to go out — not in all circumstances, but in many — to the actual substation and coordinate with customers.”

This is where Emerald AI comes in.

“For example, it’s July, we have hotter than normal weather and we lose a power plant. All of a sudden, we need 100 megawatts of reduction. Our operators send out a signal to the data center saying this is what we need. It’s surgical.”

Procos noted that Emerald AI has no direct connection to SVP’s systems or equipment.

On the data center side, Emerald AI software receives that signal and manages the power consumption of workloads designated as flexible. SVP doesn’t cut any physical connections — instead, data center managers become active partners in load management.

Unlocking Capacity by Being Flexible

“The overall vision is that any data center can opt into this program in order to unlock more power capacity, if they’re willing to be flexible,” said Emerald AI CEO Varun Sivaram. He cited a study in the journal Nature Energy reporting that tests at an Arizona data center reduced power consumption by 25% for three hours during peak demand.

“I have to give Nico and Chris Karwick, their COO, a lot of credit,” Sivaram added. “They are pioneering a really innovative program where they’re saying, we can’t fit any more data centers on our system — but if data centers are willing to be flexible, we can give them that additional capacity.”

“Imagine we can now fit more data center capacity on the SVP system,” he continued. “They can expand. They’re not constrained, because they can use more power almost all the time — and give it back when the grid needs them to.”

For now, Procos doesn’t expect Emerald AI to change SVP’s long-term infrastructure plans, but it could help the utility meet commitments already on the books. “This is something that allows us to bridge the gap until we can get to that point.”

Carolyn Schuk can be reached at carolyn@santaclaraweekly.com.